Our limited ability to store visual information over short delays is demonstrated by poor accuracy in visual working memory (VWM) tasks that require participants to match stored representations to a subsequent display to determine whether a change occurred. There are limitations both in the quantity of representations that can be held in memory, and in how well those representations match visual perception. However, VWM theorists have generally attributed errors in change detection to limitations in quantity. A method has been developed, termed mixture modeling, that models participants' responses as the sum of two independent distributions, one for guesses and one for non-guess responses. This method can estimate the quantity and fidelity of VWM representations, and determine how each property is affected by experimental manipulations. The proposed studies investigate several issues important for informing theory: a) are estimates of quantity and fidelity consistent across VWM task and object properties? b) are processes that limit quantity and fidelity shared with other capacity-limited tasks? and c) how is VWM limited under optimal conditions? These issues will be investigated by applying mixture modeling and other analysis techniques to VWM tasks under various experimental manipulations.